> On 11/17/11 15:37, Vlad Khorsun wrote:
>>> On 11/17/11 14:14, Adriano dos Santos Fernandes wrote:
>>>> On 17/11/2011 07:52, Dmitry Yemanov wrote:
>>>>> The only way to guarantee ordered writes is to flush individual pages
>>>>> and this is what FW=ON does.
>>>> Or to flush group of independent pages at once, like I proposed. I think
>>>> this can be a great improvement for FW=ON.
>>> The suggestion appears very interesting, even with average size of group
>>> == 2 engine shoudl run almost twice faster.
>> You are too optimistic :) If independent pages are not adjacent, there
>> will
>> be much less performance win.
>
> Certainly less, but how much less?
>
>> And, note, most heavy case (flush on commit)
>> is already optimized.
>
> What optimization do you talk about? As long as FW=ON, barriers=1 anyway
> forces HDD to flush write cache after each write() call.
Ooops, i mixed "batch writes via file cache" with "writes in physical
order".
Regards,
Vlad
------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure
contains a definitive record of customers, application performance,
security threats, fraudulent activity, and more. Splunk takes this
data and makes sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-novd2d
Firebird-Devel mailing list, web interface at
https://lists.sourceforge.net/lists/listinfo/firebird-devel
